Research on Intelligent Recognition Technology of Recyclable Battery Materials Combining Multi-source Perception and Deep Learning

Abstract

Traditional battery material detection methods have some problems, such as low efficiency, difficult feature extraction and multi-modal data fragmentation, which restrict the intelligent development of battery recycling. In this study, an intelligent identification technology for recycling of battery materials is proposed, which combines multi-source perception and deep learning, integrates laser-induced breakdown spectroscopy (LIBS), nano-indentation and infrared thermal imaging technology, and builds a joint detection platform to realize the synchronous characterization of material composition, structure and defects. By constructing a multi-modal feature fusion network (AMF-Net) based on attention mechanism, LIBS spectrum, nano-indentation curve and infrared thermal imaging data are fused, and multi-modal data features are mined, so as to improve the identification accuracy of battery materials. In addition, a cascade prediction network (CPN) is established to realize the accurate matching between material properties and recycling process and optimize recycling process parameters. The experimental results show that the accuracy of AMF-Net in material type identification is 92.3%, and the F1-Score is 0.91, which is significantly better than the traditional method. With the optimized recovery process parameters of CPN, the leaching rate of cobalt increased from 89.4% to 98.2%, the consumption of sulfuric acid decreased from 2.8 L/kg to 1.6 L/kg, and the reaction time was shortened from 6.5 hours to 4.2 hours, which significantly improved the recovery efficiency and reduced the resource consumption. This research provides new ideas and methods for improving battery recycling efficiency and developing green recycling technology, and promotes the development of battery recycling industry in the direction of intelligence and efficiency.